Incomex AI Portal
KB-EB3A

GPT Handoff to New Opus Session — P3D Pack1 Phase5B Start

11 min read Revision 1
handoffopusp3dpack1phase5bhybridmigration-design2026-05-11

GPT Handoff to New Opus Session — P3D Pack 1 Phase 5B Start

Date: 2026-05-11 Issuer: GPT-5.5 Thinking / Incomex Hội đồng AI Receiver: New Opus session Workstream: P3D_INFORMATION_UNIT_TEXT_AS_CODE Purpose: Start next session without search sprawl; continue from Phase 5A PASS into Phase 5B design.

0. Read this first

You are continuing P3D Pack 1 — TAC→IU migration design for the Text-as-Code / Information Unit infrastructure.

Do not restart from first principles. Do not re-audit all old phases. Do not search broadly.

Use the paths listed in §7 only. The evidence is already gathered and accepted.

1. Current accepted state

Phase 5A post-disambiguation dry-run = PASS

Accepted report:

knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase5-tac-to-iu-migration-dryrun-report.md

GPT review:

knowledge/dev/laws/dieu44-trien-khai/reviews/gpt-review-p3d-pack1-phase5a-dryrun-pass-next-phase5b-2026-05-11.md

Opus review:

knowledge/dev/laws/dieu44-trien-khai/reviews/opus-review-p3d-pack1-phase5-dryrun-pass-full-evidence-2026-05-11.md

Phase 5A evidence is complete. No more dry-run is needed unless GPT/User explicitly asks.

2. What Agent proved

Agent ran the approved Phase 5 dry-run prompt rev6 after semantic registry disambiguation.

Key accepted facts:

no_mutation_performed=true
gate0_all_passed=true
resolved_field_map_complete=true
g1_to_g11_attempted=true
old_concept_references_present=false
publication_metrics_available=true
provenance_projection_available=true
rollback_sizing_available=true
nesting_evidence_available=true
nesting_max_depth=2
address_collision_count=0
scale_brittleness_self_audit_passed=true

Source/target facts:

TAC logical units: 86
TAC unit versions: 86
TAC publications: 3
Current information_unit rows: 12
Current unit_version rows: 19
Canonical-address collisions TAC→IU: 0

Publication facts:

DIEU-28: 27 units, 7 section types, depth 2
DIEU-32: 23 units, 8 section types, depth 2
DIEU-35: 36 units, 12 section types, depth 2

Nesting facts:

roots=3
children=83
max_depth=2
parent_ref source column resolved to tac_logical_unit.parent_id
target parent_ref resolved to information_unit.parent_or_container_ref
fn_iu_create supports p_parent_ref uuid

Publication ordering:

publication_ref → tac_publication_member.publication_id
logical_unit_ref → tac_publication_member.logical_unit_id
publication_render_order → tac_publication_member.render_order
render_order is contiguous [0, n-1] per publication
no duplicates
no orphans

Hash/provenance:

TAC hash and IU hash recipe differ live.
IU must recompute fn_content_hash(body).
Original TAC hash must be preserved as provenance, not copied into IU content_hash.
Structured carrier: unit_version.content_profile jsonb.
content_profile is currently empty on 0/19 target rows populated.

Species/composition facts:

species_collection_map has no active discriminator runtime.
fn_birth_registry_auto ignores discriminator columns and NEW.unit_kind.
4 atom + 4 molecule species are PLAUSIBLE against target governance_role=observed.
No existing species is semantically locked for law_unit.
Species/composition decision must follow nesting decision.

3. GPT strategic recommendation for next step

GPT recommends using D3 HYBRID as the working migration design strategy:

D3 HYBRID = IU rows remain atomic pieces; TAC parent-child hierarchy is preserved as relationship metadata/edges, not as parent IU containment at initial migration.

Why:

  1. Keeps vector boundary clean: one IU = one text piece = one vector boundary.
  2. Preserves TAC hierarchy better than pure flattening.
  3. Avoids the mixed composition/species complexity of D1 preserve-as-IU-parent.
  4. Aligns with future typed edges / impact analysis work.
  5. Has lower blast radius than patching birth/discriminator runtime now.

This is a GPT recommendation for design. It is not an execution approval.

Recommended pilot for design planning:

DIEU-35

Reason:

highest diversity_count among the three publications; 36 units, 12 section types, depth 2, render_order valid.

Do not migrate it yet.

4. Your next task

Create Phase 5B artifacts:

P3D_PACK1_PHASE5B_HYBRID_NESTING_SPECIES_AND_PILOT_MIGRATION_DESIGN

Mode:

DESIGN + EXECUTION-PROMPT-DRAFT ONLY

No Agent execution. No DB write.

Required output 1 — Design

Create:

knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase5b-hybrid-nesting-species-pilot-migration-design.md

Required sections:

A. Executive summary
B. Accepted evidence from Phase 5A
C. Why D3 hybrid is the recommended strategy
D. D3 hybrid data model: IU atomic rows + hierarchy metadata/edges
E. Species/composition consequence of D3
F. New species / existing species decision under QT-005
G. QT-001 backfill implications for existing 12 IU birth rows
H. Pilot choice: DIEU-35 and why
I. Pilot migration algorithm draft
J. Safety gates
K. Rollback model
L. Post-implementation design requirement
M. Open questions / GPT decisions needed

Required output 2 — Execution prompt DRAFT

Create:

knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase5b-hybrid-pilot-migration-implementation-prompt-DRAFT.md

This must be a DRAFT only. It must not be dispatch-ready unless GPT later approves.

The prompt should cover a future Agent implementation for:

  1. Optional species setup under QT-005 if GPT/User chooses new species.
  2. Optional species_collection_map seed for IU if approved.
  3. Optional QT-001 backfill if approved.
  4. Pilot migration of DIEU-35 only.
  5. IU creation with atomic rows.
  6. Preserve TAC parent-child as relationship metadata/edges, not as IU parent containment, unless GPT changes D3.
  7. Preserve render_order.
  8. Recompute IU content_hash using IU recipe.
  9. Preserve original TAC content_hash in unit_version.content_profile.source_hashes.tac_v1.
  10. Exact-key rollback.
  11. Verify zero drift render after pilot.

Because it is DRAFT, it may include placeholders for GPT/User decisions, but must not contain hardcoded values as execution selectors without live verification.

Required output 3 — Report

Create:

knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase5b-design-and-draft-prompt-report.md

Required flags:

phase5b_status=PASS|PARTIAL|BLOCKED
mode=DESIGN_AND_EXECUTION_PROMPT_DRAFT_ONLY
phase5a_dryrun=PASS_ACCEPTED
recommended_nesting=D3_HYBRID
recommended_pilot=DIEU_35
migration_allowed=false
seed_allowed=false
backfill_allowed=false
execution_prompt_dispatch_allowed=false
post_implementation_design_rule_included=true|false
requires_GPT_User_review=true

5. Hard boundaries

Do not do any of the following:

No DB write.
No Agent dispatch.
No migration.
No seed.
No backfill.
No DDL.
No function/trigger patch.
No Directus/Nuxt/Qdrant changes.
No hardcoded row-count gates.
No hardcoded species/composition decision.
No hidden hardcode via column names.
No fuzzy selectors.

No code execution in this Opus step.

6. No-hardcode discipline for this session

The semantic registry has already proven its value. Continue the same principle.

Allowed scope constants:

information_unit
unit_version
law_unit
tac_logical_unit
tac_unit_version
tac_publication
tac_publication_member
DIEU-35 only as GPT-recommended pilot candidate for design planning, not as an executable selector without live re-verification.

Everything else must be either:

  1. derived live in a future executable prompt; or
  2. quoted explicitly as Phase 5A evidence; or
  3. framed as a design hypothesis pending GPT/User decision.

Do not use species names/codes as hidden selectors. Do not assume composition values from memory. Do not assume section_type list from memory. Do not assume row counts as gates.

7. Required reading only — do not search broadly

Read these files and only these unless a concrete contradiction is found:

knowledge/dev/laws/dieu44-trien-khai/reviews/gpt-review-p3d-pack1-phase5a-dryrun-pass-next-phase5b-2026-05-11.md
knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase5-tac-to-iu-migration-dryrun-report.md
knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase5-tac-to-iu-migration-design.md
knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase5a-semantic-registry-disambiguation-addendum.md
knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase4d-species-mapping-decision-memo.md
knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase4b-legal-alignment-addendum.md
knowledge/dev/laws/dieu44-trien-khai/operating-notes/design-after-repair-implementation-rule-2026-05-11.md

Do not re-search all of KB. Do not re-open old Pack 1 phases unless a contradiction appears.

8. Important context from earlier phases

  1. Phase 4 vocab prep passed. law_unit vocab exists and planner resolves it.
  2. Phase 4B/4C/4D deferred species to Phase 5 because nesting decides composition.
  3. fn_birth_registry_auto does not read discriminator columns.
  4. Discriminator-based runtime species mapping is dormant.
  5. unit_version is subordinate and has no birth rows yet.
  6. Current 12 IU birth rows have NULL species/composition.
  7. Do not repair species/backfill separately before Phase 5B design.

9. Expected Opus final response

Return only:

  1. Phase 5B design path.
  2. Phase 5B execution prompt DRAFT path.
  3. Phase 5B report path.
  4. Top 7 design choices/questions.
  5. Confirmation GPT/User review is required before any Agent dispatch.

10. Status

phase5a_dryrun=PASS_ACCEPTED
next_phase=PHASE5B_HYBRID_NESTING_SPECIES_AND_PILOT_MIGRATION_DESIGN
recommended_nesting=D3_HYBRID
recommended_pilot=DIEU_35
agent_dispatch_allowed=false
migration_allowed=false
seed_allowed=false
backfill_allowed=false
Back to Knowledge Hub knowledge/dev/laws/dieu44-trien-khai/handoffs/gpt-handoff-opus-new-session-p3d-pack1-phase5b-start-2026-05-11.md